[go: up one dir, main page]

EP3886029A1 - Procédé et système d'exécution et de recommandation d'ensemble multimodal non supervisé - Google Patents

Procédé et système d'exécution et de recommandation d'ensemble multimodal non supervisé Download PDF

Info

Publication number
EP3886029A1
EP3886029A1 EP21156595.7A EP21156595A EP3886029A1 EP 3886029 A1 EP3886029 A1 EP 3886029A1 EP 21156595 A EP21156595 A EP 21156595A EP 3886029 A1 EP3886029 A1 EP 3886029A1
Authority
EP
European Patent Office
Prior art keywords
data
features
database
interaction vector
data element
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
EP21156595.7A
Other languages
German (de)
English (en)
Inventor
Kanika Kalra
Manasi Patwardhan
Shirish Subhash Karande
Swapnil Vishveshwar Hingmire
Girish Keshav Palshikar
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tata Consultancy Services Ltd
Original Assignee
Tata Consultancy Services Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tata Consultancy Services Ltd filed Critical Tata Consultancy Services Ltd
Publication of EP3886029A1 publication Critical patent/EP3886029A1/fr
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0277—Online advertisement
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks

Definitions

  • the disclosure herein generally relates to the field of computer vision and, more particular, to a method and system for unsupervised multi-modal set completion and recommendation.
  • Online shopping is highly based on human perception of products.
  • the human perception of products depends more on semantic features of products and not just low level visual characteristics. For example, humans perceive fashion products in terms of style, art products in terms of aesthetics and music pieces in terms of genre.
  • the semantic features describes visual content of an image by correlating low level features such as color, gradient orientation with the content of an image scene.
  • semantic feature based product recommendation systems like computer vision based product recommendation systems are providing better results in online shopping domains like fashion, clothing, jewelry, furniture, beauty products, and the like.
  • a method for unsupervised multi-modal set completion and recommendation includes receiving a set of data, wherein the set of data includes at least one of, a set of images, a set of audio and a set of video captured in a plurality of modalities, wherein the each data element in the set of data is categorized. Further, the method includes computing a plurality of features for each data element by utilizing a machine learning model, wherein the plurality of features extracted at a plurality of resolutions.
  • the method includes obtaining a plurality of active features corresponding to each data element by removing irrelevant features from the plurality of features based on a predetermined threshold. Furthermore, the method includes generating a parallel corpus by arranging the plurality of active features based on the category associated with each data element. Furthermore, the method includes computing an abstract interaction vector for the set of data t by jointly modelling the parallel corpus using a probabilistic model.
  • the method includes generating a set of recommendations corresponding to the set of data based on the abstract interaction vector by utilizing a similarity metric, wherein generating the set of recommendations including: (i) computing a similarity index between the abstract interaction vector of the set of data and a pre-computed abstract interaction vector corresponding to each of a plurality of items stored in the database using the similarity metric (ii) ranking each of the plurality of items stored in the database in descending order based on the corresponding similarity index and (iii) identifying a set of items with highest similarity index from the plurality of items stored in the database for recommendation.
  • a system for unsupervised multi-modal set completion and recommendation includes at least one memory storing programmed instructions, one or more Input /Output (I/O) interfaces, and one or more hardware processors operatively coupled to the at least one memory, wherein the one or more hardware processors are configured by the programmed instructions to receive a set of data, wherein the set of data includes at least one of, a set of images, a set of audio and a set of video captured in a plurality of modalities, wherein the each data element in the set of data is categorized.
  • I/O Input /Output
  • the one or more hardware processors are configured by the programmed instructions to compute a plurality of features for each data element by utilizing a machine learning model, wherein the plurality of features extracted at a plurality of resolutions. Furthermore, the one or more hardware processors are configured by the programmed instructions to obtain a plurality of active features corresponding to each data element by removing irrelevant features from the plurality of features based on a predetermined threshold. Furthermore, the one or more hardware processors are configured by the programmed instructions to generate a parallel corpus by arranging the plurality of active features based on the category associated with each data element. Furthermore, the one or more hardware processors are configured by the programmed instructions to compute an abstract interaction vector for the set of data t by jointly modelling the parallel corpus using a probabilistic model.
  • the one or more hardware processors are configured by the programmed instructions to generate a set of recommendations corresponding to the set of data based on the abstract interaction vector by utilizing a similarity metric, wherein generating the set of recommendations including: (i) computing a similarity index between the abstract interaction vector of the set of data and a pre-computed abstract interaction vector corresponding to each of a plurality of items stored in the database using the similarity metric (ii) ranking each of the plurality of items stored in the database in descending order based on the corresponding similarity index and (iii) identifying a set of items with highest similarity index from the plurality of items stored in the database for recommendation.
  • a computer program product including a non-transitory computer-readable medium having embodied therein a computer program for method and system for unsupervised multi-modal set completion and recommendation.
  • the computer readable program when executed on a computing device, causes the computing device to receive a set of data, wherein the set of data includes at least one of, a set of images, a set of audio and a set of video captured in a plurality of modalities, wherein the each data element in the set of data is categorized.
  • the computer readable program when executed on a computing device, causes the computing device to compute a plurality of features for each data element by utilizing a machine learning model, wherein the plurality of features extracted at a plurality of resolutions.
  • the computer readable program when executed on a computing device, causes the computing device to obtain a plurality of active features corresponding to each data element by removing irrelevant features from the plurality of features based on a predetermined threshold. Furthermore, the computer readable program, when executed on a computing device, causes the computing device to generate a parallel corpus by arranging the plurality of active features based on the category associated with each data element. Furthermore, the computer readable program, when executed on a computing device, causes the computing device to compute an abstract interaction vector for the set of data t by jointly modelling the parallel corpus using a probabilistic model.
  • the computer readable program when executed on a computing device, causes the computing device to generate a set of recommendations corresponding to the set of data based on the abstract interaction vector by utilizing a similarity metric, wherein generating the set of recommendations including: (i) computing a similarity index between the abstract interaction vector of the set of data and a pre-computed abstract interaction vector corresponding to each of a plurality of items stored in the database using the similarity metric (ii) ranking each of the plurality of items stored in the database in descending order based on the corresponding similarity index and (iii) identifying a set of items with highest similarity index from the plurality of items stored in the database for recommendation.
  • Embodiments herein provide a method and system for unsupervised multi-modal set completion and recommendation to perform set completion and recommendation in accurate manner.
  • the present disclosure focuses on problem of proposing a common abstract representation for a set of data.
  • the set of data including at least one of a set of images, a set of videos and a set of audio are given as input to the system.
  • a plurality of features are extracted from the set of data at a plurality of resolution levels and the plurality of features are arranged as parallel corpus based on a category associated with each data from the set of data.
  • an abstract interaction vector is computed for each element of the set of data using the parallel corpus.
  • the set of recommendations are identified by comparing the abstract interaction vector associated with the set of data with an abstract interaction vector associated with each of a plurality of items stored in the database by utilizing a similarity metric.
  • the present disclosure is useful in set recommendation or set completion problem where suggesting a set of products which go along well together is required.
  • the method for unsupervised multi-modal set completion and recommendation also addresses the problem of the how visual appearance of the set of items are perceived by a user in order to provide effective online set recommendation.
  • the present disclosure allows online outfit recommendation with a coherent fashion style, say formal, or photos assortment for a collage which leads to high aesthetic value, etc.
  • the present disclosure identifies a coherent general theme that allows to understand image compatibility beyond visual similarity for product recommendation.
  • the present disclosure solves the problem of computing a common representation learning by using only images and categories associated with the images without requiring any external data labels as in the case of supervised learning.
  • the present disclosure does not require semantic attributes, comments from set posts on social media and the like.
  • FIG. 1 through 3 where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments and these embodiments are described in the context of the following exemplary system and/or method.
  • FIG. 1 is a functional block diagram of a system 100 for unsupervised multi-modal set completion and recommendation, according to some embodiments of the present disclosure.
  • the system 100 includes or is otherwise in communication with hardware processors 102, at least one memory such as a memory 104, an I/O interface 112.
  • the hardware processors 102, memory 104, and the Input /Output (I/O) interface 112 may be coupled by a system bus such as a system bus 108 or a similar mechanism.
  • the hardware processors 102 can be one or more hardware processors.
  • the I/O interface 112 may include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like.
  • the I/O interface 112 may include a variety of software and hardware interfaces, for example, interfaces for peripheral device(s), such as a keyboard, a mouse, an external memory, a printer and the like. Further, the interface 112 may enable the system 100 to communicate with other devices, such as web servers and external databases.
  • the I/O interface 112 can facilitate multiple communications within a wide variety of networks and protocol types, including wired networks, for example, local area network (LAN), cable, etc., and wireless networks, such as Wireless LAN (WLAN), cellular, or satellite.
  • the I/O interface 112 may include one or more ports for connecting a number of computing systems with one another or to another server computer.
  • the I/O interface 112 may include one or more ports for connecting a number of devices to one another or to another server.
  • the one or more hardware processors 102 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, and/or any devices that manipulate signals based on operational instructions.
  • the one or more hardware processors 102 is configured to fetch and execute computer-readable instructions stored in the memory 104.
  • the memory 104 may include any computer-readable medium known in the art including, for example, volatile memory, such as static random access memory (SRAM) and dynamic random access memory (DRAM), and/or non-volatile memory, such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes.
  • volatile memory such as static random access memory (SRAM) and dynamic random access memory (DRAM)
  • non-volatile memory such as read only memory (ROM), erasable programmable ROM, flash memories, hard disks, optical disks, and magnetic tapes.
  • the memory 104 includes a plurality of modules 106 and a data analysis unit 114.
  • the memory 104 also includes a data repository 110 for storing data processed, received, and generated by the plurality of modules 106 and the data analysis unit 114.
  • the plurality of modules 106 include programs or coded instructions that supplement applications or functions performed by the system 100 for unsupervised multi-modal set completion and recommendation.
  • the plurality of modules 106 can include routines, programs, objects, components, and data structures, which performs particular tasks or implement particular abstract data types.
  • the plurality of modules 106 may also be used as, signal processor(s), state machine(s), logic circuitries, and/or any other device or component that manipulates signals based on operational instructions. Further, the plurality of modules 106 can be used by hardware, by computer-readable instructions executed by a processing unit, or by a combination thereof.
  • the plurality of modules 106 can include various sub-modules (not shown).
  • the plurality of modules 106 may include computer-readable instructions that supplement applications or functions performed by the system 100 for unsupervised multi-modal set completion and recommendation.
  • the data repository 110 may include a plurality of abstracted piece of code for refinement and data that is processed, received, or generated as a result of the execution of the plurality of modules in the module(s) 106 and the modules associated with the data analysis unit 114.
  • the data repository or database may also include a plurality of items including image data, video data, text data and audio data.
  • the data repository 110 is shown internal to the system 100, it will be noted that, in alternate embodiments, the data repository 110 can also be implemented external to the computing device 100, where the data repository 110 may be stored within a database (not shown in FIG. 1 ) communicatively coupled to the system 100.
  • the data contained within such external database may be periodically updated. For example, new data may be added into the database (not shown in FIG. 1 ) and/or existing data may be modified and/or non-useful data may be deleted from the database (not shown in FIG. 1 ).
  • the data may be stored in an external system, such as a Lightweight Directory Access Protocol (LDAP) directory and a Relational Database Management System (RDBMS).
  • LDAP Lightweight Directory Access Protocol
  • RDBMS Relational Database Management System
  • FIG. 2 is exemplary flow diagram for a processor implemented method for unsupervised multi-modal set completion and recommendation implemented by the system of FIG. 1 , according to some embodiments of the present disclosure.
  • the system 100 comprises one or more data storage devices or the memory 104 operatively coupled to the one or more hardware processor(s) 102 and is configured to store instructions for execution of steps of the method 200 by the one or more hardware processors 102.
  • the steps of the method 200 of the present disclosure will now be explained with reference to the components or blocks of the system 100 as depicted in FIG. 1 and the steps of flow diagram as depicted in FIG. 2 .
  • the method 200 may be described in the general context of computer executable instructions.
  • computer executable instructions can include routines, programs, objects, components, data structures, procedures, modules, functions, etc., that perform particular functions or implement particular abstract data types.
  • the method 200 may also be practiced in a distributed computing environment where functions are performed by remote processing devices that are linked through a communication network.
  • the order in which the method 200 is described is not intended to be construed as a limitation, and any number of the described method blocks can be combined in any order to implement the method 200, or an alternative method.
  • the method 200 can be implemented in any suitable hardware, software, firmware, or combination thereof.
  • the one or more hardware processors (102) receive a set of data, wherein the set of data includes at least one of, a set of images, a set of audio and a set of video captured in a plurality of modalities, wherein the each data element in the set of data is categorized.
  • the set of data can also include textual data.
  • the one or more hardware processors (102) compute a plurality of features for each data element from the set of data by utilizing a machine learning model. The plurality of features extracted at a plurality of resolutions.
  • the one or more hardware processors (102) obtain a plurality of active features corresponding to each data element from the set of data by removing irrelevant features from the plurality of features based on a predetermined threshold.
  • the one or more hardware processors (102) generate a parallel corpus by arranging the plurality of active features based on a category associated with each data element from the set of data.
  • the parallel corpus includes a plurality of active features corresponding to each category of data element.
  • the one or more hardware processors (102) computes an abstract interaction vector for the set of data using the parallel corpus by jointly modelling the parallel corpus using a probabilistic model.
  • the joint modelling projects a lower dimensional space such that probability of joint occurrences of the active features of each corpus is maximized by the probabilistic model.
  • the probabilistic model can be an extension of Latent Dirichlet Allocation (LDA) or Auto encoders capable of receiving a set as input.
  • the database includes the plurality of items and the abstract interaction vector corresponding to each of the plurality of items are pre-computed.
  • the one or more hardware processors (102) generates a set of recommendations corresponding to the set of data based on the abstract interaction vector by utilizing a similarity metric, wherein the similarity metric includes one of a Euclidean distance and cosine similarity.
  • the method of generating the set of recommendations includes the following steps: (i) computing a similarity index between the abstract interaction vector of the set of data and the pre-computed abstract interaction vector corresponding to each of the plurality of items stored in the database using the similarity metric (ii) ranking each of the plurality of items stored in the database in descending order based on the corresponding similarity index and (iii) identifying a set of items with highest similarity index from the plurality of items stored in the database for recommendation.
  • FIG. 3 illustrates a functional block diagram of the system of FIG. 1 for unsupervised multi-modal set completion and recommendation, in accordance with some embodiments of the present disclosure.
  • the functional block diagram includes a module for feature extraction using a machine learning model 302, a module for corpus generation 304, a module for abstract interaction vector computation 306 and a module for recommending the set of items which go well with each other 308.
  • the modules 302, 304, 306 and 308 are present inside the data analysis unit 114.
  • the module 302 extracts the plurality of features corresponding to each data from the data set at a plurality resolution levels or in a plurality of granular levels by using the machine learning model.
  • the machine learning model may be pre-trained with ImageNet data set and not with domain specific images or data and hence unsupervised.
  • the machine learning model includes a plurality of filters to generate the plurality of features at the plurality of granular levels. Each filter from the plurality of filters can compute a feature.
  • the plurality of features in different resolution/granularity are obtained using the feature extraction module 302.
  • the plurality of active features are obtained by removing irrelevant features from the plurality of features by using the predetermined threshold.
  • the predetermined threshold is 0.8.
  • the module 304 generates the corpus corresponding to each category of data from the set of data simultaneously.
  • Each corpus includes the active features corresponding to each category of data from the set of data.
  • the module 306 computes abstract interaction vector of the set of data based on the parallel corpus by jointly modelling each corpus using a probabilistic model.
  • the joint modelling projects the lower dimensional space such that probability of joint occurrences of the active features of each corpus is maximized by the probabilistic model.
  • the module 308 recommends the set of items which go well with each other.
  • the items are recommended based on the comparison between abstract interaction vector associated with the set of data and the abstract interaction vectors corresponding to each of the plurality of items stored in the database.
  • the comparison is done based on the similarity index computed using the similarity metric.
  • the data analysis unit 114 executed by the one or more processors of the system 100, receives the set of data, wherein the set of data includes at least one of, the set of images, the set of audio and the set of video captured in a plurality of modalities, wherein the each data element in the set of data is categorized.
  • the set of images can be a set of fashion items which forms an outfit representing retro style - Polka dotted top, bell bottom pants, red bag.
  • the set of audio can be a set of audio clips which represents a genre say pop, wherein the audio clips are taken from the same or multiple songs.
  • the set of video can be video clips representing romance as an abstract concept, the video clips could be from the same movie or different movie, where characters may be holding hands, looking into each other eyes etc.
  • the input set of data can also include textual data.
  • the data analysis unit 114 executed by one or more processors of the system 100, compute the plurality of features for each data element from the set of data by utilizing the machine learning model, wherein the plurality of features are extracted at the plurality of resolutions.
  • the plurality of features can be computed by using a plurality of techniques.
  • the plurality of features can be the outputs of convolution layers of a pre-trained Convolution Neural Network (CNN) formally known as filter maps, which captures characteristics of the input data at distinct resolution depending on the layer from which the output is extracted.
  • CNN may be a generic pre-trained network with ImageNet data set and not with domain specific images or data.
  • the plurality of features can be computed by using other computer vision techniques including HOG (Histogram Oriented Gradients), SIFT (Scale Invariant Feature Transform) for image patches at different resolution or for a full image.
  • HOG Heistogram Oriented Gradients
  • SIFT Scale Invariant Feature Transform
  • the data analysis unit 114 executed by one or more processors of the system 100, obtains the plurality of active features corresponding to each data element from the set of data by removing irrelevant features from the plurality of features based on the predetermined threshold.
  • the plurality of features from each layer of the pre-trained CNN are extracted from filter map.
  • the size of the filters depends on the pre-trained CNN being used. Since total number of filters obtained is very high, the present disclosure uses the predefined threshold to understand which filter could be irrelevant based on some heuristics. The heuristic could be different for different layers. For example, the heuristic is "if 3/4 of the filter map grid has values greater than 0.8 (predefined filter threshold), the filters are considered to be active and provide active features.
  • the data analysis unit 114 executed by one or more processors of the system 100, generates the parallel corpus by arranging the plurality of active features based on the category associated with each data element from the set of data.
  • the parallel corpus includes the plurality of active features corresponding to each category of data from the set of data.
  • the plurality of active features are computed for each data element of the set of data.
  • a set S having 4 elements E1,E2,E3,E4 and each element belong to categories C1,C2,C3,C4 respectively.
  • Features are computed for each element E1,E2,E3,E4 and the features obtained are F1,F2,F3,F4 respectively.
  • the organization of the features to categories and storing in the data structure for multiple set is called corpus generation.
  • the corpus is referred to as parallel corpus as each row corresponds to a set and each element is a set are organized in parallel.
  • the corpus for the above example is as shown in table 1.
  • Each row is a set.
  • Each set (row) in the table 1 includes features from 4 categories (described in columns).
  • F11 indicates feature 1 of set 1
  • F12 indicates feature 2 of set 1
  • F32 indicates feature 2 of set 3 and the like.
  • the data analysis unit 114 executed by one or more processors of the system 100, computes the abstract interaction vector for the set of data by jointly modelling the parallel corpus using the probabilistic model like Latent Dirichlet Allocation (LDA).
  • LDA Latent Dirichlet Allocation
  • the joint modelling projects a lower dimensional space such that probability of joint occurrences of the active features of each corpus is maximized by the probabilistic model.
  • the joint modelling can happen using any probabilistic graphical model such as an extension of LDA or auto encoders (any model which can learn abstract latent representation) which takes a set as an input instead of one single item.
  • abstract interaction vector or theme can be computed using auto encoder where the input is reconstructed by projecting it to lower dimension latent space. It is known to discover the commonalities of the input features are correlated.
  • abstract interaction vector or theme can be computed using topic modeling approach.
  • the abstract interaction vector is alternatively represented as theme or latent representation throughout the document.
  • Each category is treated as a language and modelled as a multilingual topic modeling problem.
  • This model learns a shared topic document distribution across the products within a set by learning the product feature distribution.
  • the topic modelling approach has been used in literature to jointly model multiple modalities.
  • the comparable corpus or corpus contains one language as text and other as image or by modelling text description of different product categories as multiple languages.
  • image features of products belonging to different categories are used as multiple languages.
  • the topic model trained on set product feature indexes data can be used to obtain topic distribution.
  • the topic distribution can be extracted even for product set with missing product language information. This topic distribution represent the general theme.
  • the data analysis unit 114 executed by one or more processors of the system 100, generates the set of recommendations corresponding to the set of data based on the abstract interaction vector by utilizing the similarity metric, wherein the similarity metric includes one of the Euclidean distance and the cosine similarity.
  • the method of generating the set of recommendations includes the following steps: (i) computing the similarity index between the abstract interaction vector corresponding to the set of data and the pre-computed abstract interaction vector corresponding to each of the plurality of items stored in the database using the similarity metric (ii) ranking each of the plurality of items stored in the database in descending order based on the corresponding similarity index and (iii) identifying the set of items with highest similarity index from the plurality of items stored in the database for recommendation.
  • the present disclosure can be used in outfit search. For example, if there is a set of fashion items representing a retro style and there is a need to search a similar retro style outfit in the database for recommendation.
  • the present disclosure can compute latent representation/ abstract interaction vectors of all items in the database and the input set of data. Further, the similarity index between the latent representation of the input set of data and the latent representation of the plurality of items stored in the database are computed and compared. The plurality of items in the database are ranked based on the similarity index. The set of item with highest similarity is identified from the plurality of items and recommended.
  • the latent representation is computed for the original input set including the top outfit and the bag.
  • a plurality of sets are formed by concatenating the original input set of items in the set and each missing element in the choice list or the plurality of items (which is stored in the database).
  • the latent representation is computed for the formed sets.
  • similarity index is computed between the original set representation and sets formed with choice as an element in the set.
  • the bottom outfit choices are ranked and the top k elements from missing element category are recommended.
  • the present disclosure can be utilized in subscription box services where set recommendation is of huge importance.
  • a story book subscription service suggests set of story books which can be recommended together, or a beauty box subscription service which recommends set of beauty products to women, etc.
  • For online subscription outfit recommendation service which recommends a set of clothing and accessary images.
  • the present disclosure can learn common abstract representation for a product set using only product images without external labelled data and hence it is unsupervised.
  • the present disclosure would be capable of extracting style representation for a complete outfit (set) given the set of images of individual elements such as upper and lower body garments, bags, shoes, jewelry, etc., or given a set of images of products which would compose a display for a retail store, our system would provide an aesthetic representation of the display consisting of these set of products.
  • the embodiments of present disclosure herein address unresolved problem of unsupervised multi-modal set recommendation.
  • the present disclosure learns abstract common representation or general theme of set of images (products) by modelling their interactions jointly at different resolution levels.
  • the present disclosure does not require any labelled data and thus can work in a completely unsupervised fashion.
  • the present disclosure obtains the general theme or the common representation for the product set which exists in a latent space. This representation is more semantic and abstract.
  • the present disclosure implements a model which is robust enough to accommodate the common concepts which are distributed among the set of images at different granularities and resolution levels leading to a common representation that can be learned together for the overall set.
  • the hardware device can be any kind of device which can be programmed including e.g. any kind of computer like a server or a personal computer, or the like, or any combination thereof.
  • the device may also include means which could be e.g. hardware means like e.g. an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of hardware and software means, e.g.
  • ASIC application-specific integrated circuit
  • FPGA field-programmable gate array
  • the means can include both hardware means and software means.
  • the method embodiments described herein could be implemented in hardware and software.
  • the device may also include software means.
  • the embodiments may be implemented on different hardware devices, e.g. using a plurality of CPUs, GPUs and edge computing devices.
  • the embodiments herein can comprise hardware and software elements.
  • the embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc.
  • the functions performed by various modules described herein may be implemented in other modules or combinations of other modules.
  • a computer-usable or computer readable medium can be any apparatus that can comprise, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
  • a computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored.
  • a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein.
  • the term "computer-readable medium” should be understood to include tangible items and exclude carrier waves and transient signals, i.e. non-transitory. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, nonvolatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and any other known physical storage media.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Business, Economics & Management (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computational Mathematics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Algebra (AREA)
  • Mathematical Optimization (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Pure & Applied Mathematics (AREA)
  • Game Theory and Decision Science (AREA)
  • Mathematical Analysis (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Image Analysis (AREA)
EP21156595.7A 2020-03-26 2021-02-11 Procédé et système d'exécution et de recommandation d'ensemble multimodal non supervisé Withdrawn EP3886029A1 (fr)

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
IN202021013242 2020-03-26

Publications (1)

Publication Number Publication Date
EP3886029A1 true EP3886029A1 (fr) 2021-09-29

Family

ID=74591880

Family Applications (1)

Application Number Title Priority Date Filing Date
EP21156595.7A Withdrawn EP3886029A1 (fr) 2020-03-26 2021-02-11 Procédé et système d'exécution et de recommandation d'ensemble multimodal non supervisé

Country Status (2)

Country Link
US (1) US20210304072A1 (fr)
EP (1) EP3886029A1 (fr)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11467851B1 (en) * 2019-11-21 2022-10-11 Synopsys, Inc. Machine learning (ML)-based static verification for derived hardware-design elements
CN114385873B (zh) * 2021-12-31 2024-08-13 北京师范大学 基于长文本语义一致的家具场景重建与推荐方法
CN114692778B (zh) * 2022-04-13 2023-07-25 北京百度网讯科技有限公司 用于智能巡检的多模态样本集生成方法、训练方法及装置

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180052928A1 (en) * 2016-08-16 2018-02-22 Ebay Inc. Semantic forward search indexing of publication corpus

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10394960B2 (en) * 2016-12-21 2019-08-27 Facebook, Inc. Transliteration decoding using a tree structure
US10140289B2 (en) * 2017-03-29 2018-11-27 Galisteo Consulting Group, Inc. Identifying propaganda in global social media
WO2019147767A1 (fr) * 2018-01-24 2019-08-01 Rensselaer Polytechnic Institute Autocodeur convolutif 3-d pour tomographie assistée par ordinateur à faible dose par l'intermédiaire d'un apprentissage de transfert à partir d'un réseau entraîné en 2-d
US20210342496A1 (en) * 2018-11-26 2021-11-04 Hewlett-Packard Development Company, L.P. Geometry-aware interactive design
US11507845B2 (en) * 2018-12-07 2022-11-22 Accenture Global Solutions Limited Hybrid model for data auditing
US12131365B2 (en) * 2019-03-25 2024-10-29 The Board Of Trustees Of The University Of Illinois Search engine use of neural network regressor for multi-modal item recommendations based on visual semantic embeddings
US11593705B1 (en) * 2019-06-28 2023-02-28 Amazon Technologies, Inc. Feature engineering pipeline generation for machine learning using decoupled dataset analysis and interpretation
US11763091B2 (en) * 2020-02-25 2023-09-19 Palo Alto Networks, Inc. Automated content tagging with latent dirichlet allocation of contextual word embeddings

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180052928A1 (en) * 2016-08-16 2018-02-22 Ebay Inc. Semantic forward search indexing of publication corpus

Also Published As

Publication number Publication date
US20210304072A1 (en) 2021-09-30

Similar Documents

Publication Publication Date Title
De Divitiis et al. Disentangling features for fashion recommendation
Ullman et al. Atoms of recognition in human and computer vision
KR102741221B1 (ko) 스트리밍 비디오 내의 객체를 검출하고, 필터링하고 식별하기 위한 방법 및 장치
CN111784455A (zh) 一种物品推荐方法及推荐设备
Mohammadi et al. Smart fashion: a review of AI applications in the Fashion & Apparel Industry
US20230214895A1 (en) Methods and systems for product discovery in user generated content
US20130036121A1 (en) System and method for recommending blog
CN110909536A (zh) 用于自动生成产品的文章的系统和方法
US20180121798A1 (en) Recommender system
EP3886029A1 (fr) Procédé et système d'exécution et de recommandation d'ensemble multimodal non supervisé
CN107632984A (zh) 一种聚类数据表的展现方法、装置和系统
Feng et al. Interpretable partitioned embedding for customized multi-item fashion outfit composition
CN107835994A (zh) 通过图像的任务聚焦搜索
US20200226168A1 (en) Methods and systems for optimizing display of user content
US20230030560A1 (en) Methods and systems for tagged image generation
US20250298837A1 (en) Methods and systems for dynamic content provisioning
Mehmood et al. Effect of complementary visual words versus complementary features on clustering for effective content-based image search
CN115114519A (zh) 基于人工智能的推荐方法、装置、电子设备及存储介质
KR102119518B1 (ko) 인공지능을 이용하여 생성되는 스타일 공간에 기반한 상품 추천 방법 및 시스템
Chi et al. UbiShop: Commercial item recommendation using visual part-based object representation
Stacchio et al. Toward a holistic approach to the socio-historical analysis of vernacular photos
CN113763084B (zh) 产品推荐的处理方法、装置、设备及存储介质
Lei et al. A new clothing image retrieval algorithm based on sketch component segmentation in mobile visual sensors
Yuan et al. Matching recommendations based on siamese network and metric learning
CN114780736B (zh) 客服知识库的构建方法和装置

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE APPLICATION HAS BEEN PUBLISHED

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE

17P Request for examination filed

Effective date: 20220329

RBV Designated contracting states (corrected)

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: EXAMINATION IS IN PROGRESS

17Q First examination report despatched

Effective date: 20231120

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE APPLICATION HAS BEEN WITHDRAWN

18W Application withdrawn

Effective date: 20250623